I start to work with a dataset with more than 3000 samples, and realize that ACF plot does not make sense with respect of confidence intervals, see figure.

enter image description here

My purpose is to fit an ARIMA on this, however, I'm not sure if it's a good choice because of this large sample size, as @IrishStat answer alerts. Does anyone recommend an approach to handle this ACF plot to model the ARIMA?

EDIT: In acf, the first lag inside interval confidence is in something about 400.

  • $\begingroup$ What does the original time series look like? What is it a time series of? What's the time interval here? $\endgroup$ – The Laconic Apr 6 '17 at 2:58

From the ACF plot, it seems there is periodic behind it , try the seasonal difference.

If possible, do the unit root test to make sure the series is stationary before chooses the parameters for arima .


In case of a large sample you don't need to do ACF, you can use spectral analysis with periodograms, FFT etc. ACF and ARIMA is a poor man's spectral analysis when the sample is small. Whether 3,000 is a large sample is a different question


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.